/ctrl

Primary LanguageJupyter NotebookMIT LicenseMIT

CTRL: Clustering Training Losses for Label Error Detection

This repo contains the implementation of CTRL and files for running experiments reported in the paper.

  • clustering.py computes masks using CTRL.
  • confident_learning.py computes masks using Confident Learning methods.
  • cifar10 contains files for CIFAR10 experiments.
  • cifar100 contains instructions of how to modify files in cifar10 to conduct experiments for CIFAR100.
  • tabular contains files for tabular datasets experiments.

Please cite our paper if you use this package.